Fonctions de déformation image produit tensoriel généralisées Generalized Tensor-Product Warps

The inter-image flow field is often modeled by some parametric warp function. Nearly all warps in the literature are based on a linear combination of control points, as for example the Free-Form Deformation (FFD) warp which uses the tensor-product of B-splines. It has been recently shown that the FFD warp models the affine projection of a deforming surface. This is also the case for all tensorproduct warps and this limits the extent of deformations that can be modeled by these warps. We present the Generalized Tensor-Product (GTP) warps. These model the affine and perspective projection of a rigid and a deforming surface in a generic manner. They include the FFD warp and the recent NURBS warp as special cases. We also propose a new kind of warp that is simpler than the perspective warp using an affine function to model the surface’s depth. Experimental results are reported for simulated and real data, showing how our GTP warps improve on existing ones.